A carregar...
Publicação
Identificação de operações de limpeza em faixas de gestão de combustível de incêndios ao redor de habitações e localidades
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 8.07 MB | Adobe PDF |
Orientador(es)
Resumo(s)
As florestas portuguesas são todos os anos ameaçadas por incêndios florestais. Para mi-
nimizar esse flagelo, existem planos de defesa com o objectivo de diminuir o risco de
incêndio e limitar a sua propagação. Nesses planos incluem-se as faixas de gestão de com-
bustível (FGCI), zonas onde a presença de vegetação deve ser reduzida ou inexistente.
Nesta dissertação é proposto o uso de imagens de satélite, denominadamente da mis-
são Sentinel-2 para monitorizar o estado das FGCI, identificando os instantes de interven-
ção sobre as FGCI, tal como a sua ausência.
O uso de técnicas de aprendizagem automática como Random Forests são aplicadas
para distinguir um instante num contexto normal, e num contexto pós-intervenção. São
usados os dados de referência fornecidos e complementados por um processo manual
de marcação de séries para treinar os modelos. O contexto temporal fornece informação
importante para apoiar esta classificação.
Foi explorada também a hipótese de gerar automaticamente as FGCI, com base na
deteção de estruturas permanentes, de forma a identificar faixas adicionais que não estão
mapeadas nos planos de defesa.
Every year, the Portuguese forests are threatened by wild fires. To minimize that danger, there are mitigation plans put in place with the intent of lowering the fire risk and limiting its spread. Those plans include the definition of Fire Management Zones (FMZ), zones in which the vegetation should be routinely trimmed, or removed. In this thesis it is proposed the usage of satellite imagery, namely from the Sentinel-2 mission, to monitor the state of the FMZs, identifying the moments where an intervention occured, as well as the periods without any action, such as when the vegetation cover evolves naturally. The usage of machine learning techniques like Random Forests are applied to dis- tinguish a normal context in an FMZ’s vegetation cycle and a moment that follows an intervention. These techniques use the ground truth data that was provided, comple- mented by a manual effort of time series tagging to be able to train these models. The temporal context provides valuable information to support the classification. The possibility of automatically generating the FMZ’s was also explored, relying on the detection of permanent artifical strutctures to generate them. This new map of au- tomatic FMZ’s can be used to detect aditional magament zones that are not currently defined in the oficial plans.
Every year, the Portuguese forests are threatened by wild fires. To minimize that danger, there are mitigation plans put in place with the intent of lowering the fire risk and limiting its spread. Those plans include the definition of Fire Management Zones (FMZ), zones in which the vegetation should be routinely trimmed, or removed. In this thesis it is proposed the usage of satellite imagery, namely from the Sentinel-2 mission, to monitor the state of the FMZs, identifying the moments where an intervention occured, as well as the periods without any action, such as when the vegetation cover evolves naturally. The usage of machine learning techniques like Random Forests are applied to dis- tinguish a normal context in an FMZ’s vegetation cycle and a moment that follows an intervention. These techniques use the ground truth data that was provided, comple- mented by a manual effort of time series tagging to be able to train these models. The temporal context provides valuable information to support the classification. The possibility of automatically generating the FMZ’s was also explored, relying on the detection of permanent artifical strutctures to generate them. This new map of au- tomatic FMZ’s can be used to detect aditional magament zones that are not currently defined in the oficial plans.
Descrição
Palavras-chave
Deteção Remota Aprendizagem Automática Séries Temporais Faixas de Gestão de Combustível Sentinel